By: Dauda Lawal – Web Developer
Safety remains one of the most important operational priorities across the oil and gas industry. Processing plants, flow stations, pipeline facilities and petroleum storage terminals operate in environments where a single safety failure can lead to serious injuries, equipment damage, environmental incidents and regulatory penalties.
However, maintaining consistent safety oversight across multiple facilities is not always straightforward. Health, Safety and Environment (HSE) teams must monitor worker behaviour, enforce personal protective equipment (PPE) requirements, control access to restricted areas, identify visible hazards and maintain accurate incident records.
Traditional inspection routines can only cover a limited portion of daily operations. Even when facilities have extensive CCTV infrastructure, cameras often serve primarily as security and post-incident investigation tools rather than active safety-monitoring systems.
To address these challenges, Arthurite Integrated implemented an AI-powered HSE Monitoring and Incident Management Platform on Amazon Web Services (AWS) for Northbridge Energy Limited, an integrated Nigerian oil and gas company.
The solution transforms existing CCTV infrastructure into an intelligent safety-monitoring system. By combining computer vision, real-time video processing, machine learning, generative AI and automated incident workflows, the platform helps Northbridge identify potential hazards, notify the appropriate personnel and improve the preparation of safety and regulatory reports.
The initial deployment covers 64 cameras across six oil and gas facilities, including processing plants, flow stations and petroleum product storage terminals.
This article explores the operational challenges behind the engagement, the AWS architecture implemented, the engineering capabilities powering the platform and the business outcomes Northbridge aims to achieve.
The Challenge: Why Traditional HSE Monitoring Was No Longer Enough
1. Limited Visibility Across Operating Hours
Northbridge Energy operates facilities where personnel work around rotating equipment, pressurized systems, petroleum products and other potentially hazardous assets.
Before the engagement, the company’s HSE team relied heavily on physical inspections and periodic compliance checks. Approximately 30 manual inspections took place each month, covering less than 2% of total working hours.
Although these inspections helped identify some safety issues, they could not provide continuous visibility across all operating areas.
Consequently, PPE violations, unauthorized access to restricted areas and other unsafe activities could occur without immediate detection.
For example, a worker entering an operational area without the required protective equipment might remain unnoticed until the next inspection. Similarly, unauthorized entry into a permit-controlled zone could only become visible during a subsequent audit or incident investigation.
This created a significant gap between the occurrence of unsafe behaviour and the ability of the HSE team to respond.
2. Delayed Detection of Leaks, Smoke and Spills
Oil and gas facilities must respond quickly to visible leaks, smoke, flames and liquid spills. Delayed detection can increase the severity of an incident, expose personnel to hazardous conditions and create additional environmental and financial consequences.
Northbridge’s existing processes depended on physical rounds, manual observations and reports from personnel working within the facilities.
The baseline documented for the engagement estimated that leak and related hazard detection-to-response took approximately 4.5 hours on average.
That delay created unnecessary exposure. A small spill or visible leak could develop into a more serious operational incident before the appropriate personnel received an alert.
3. Underutilized CCTV Infrastructure
Northbridge had approximately 210 installed CCTV cameras across its facilities. However, these cameras primarily supported security monitoring and retrospective investigations.
The cameras were not being used to systematically identify PPE violations, restricted-zone entry, smoke, flames and spills.
This meant the organization already had a valuable source of operational visibility but lacked the intelligent processing capabilities required to turn video footage into actionable safety information.
Rather than replacing the existing infrastructure, Arthurite designed a solution that reuses the available cameras and extends their capabilities through AI-powered video analytics.
4. Slow Incident Investigation and Regulatory Reporting
Incident management involves more than detecting an event. HSE teams must collect evidence, establish timelines, document observations, prepare incident summaries and produce regulatory reports.
Northbridge’s manual reporting process required approximately six working days to prepare incident reports and around three working days to prepare regulatory report drafts.
This administrative workload reduced the time HSE personnel could dedicate to preventive safety activities.
The company therefore needed a platform that could capture incident evidence, organize relevant information and accelerate report preparation without removing human accountability from safety decisions.
The Business Cost of Reactive HSE Management
The workbook estimates Northbridge’s annual cost of reactive HSE management at approximately ₦418 million.
| Cost category | Estimated annual impact |
|---|---|
| Injury-related costs and lost time | ₦120 million |
| Regulatory fines and penalty exposure | ₦150 million |
| Product loss and late leak-response costs | ₦65 million |
| Manual inspection and reporting effort | ₦38 million |
| Insurance and reputational impact | ₦45 million |
| Total estimated annual impact | ₦418 million |
These figures demonstrate why improving safety visibility is not simply a technology upgrade. It is an operational priority with direct implications for personnel protection, regulatory compliance, asset preservation and business continuity.
The AWS Solution: An Intelligent HSE Monitoring and Incident Management Platform
Arthurite Integrated implemented an AWS-based architecture that connects existing camera infrastructure to real-time video analytics, machine learning models, automated alerting and AI-assisted reporting.
The solution monitors 64 selected cameras across six Northbridge facilities.
At each facility, camera gateways collect video streams from existing CCTV systems. The gateways apply local filtering, reduce video resolution and transmit the relevant footage through redundant AWS Site-to-Site VPN connections.
Amazon Kinesis Video Streams provides the video ingestion layer, while AWS Lambda and container-based consumers extract selected frames for analysis.
Amazon SageMaker hosts custom computer-vision models that identify PPE violations, restricted-zone entry, smoke, flames and spills. Amazon Rekognition provides additional verification for flagged frames, helping reduce false alerts.
When the platform confirms a potential safety event, Amazon EventBridge, AWS Lambda and AWS Step Functions coordinate the incident workflow. Amazon SNS delivers notifications to the appropriate personnel.
For incident documentation, Amazon Bedrock generates structured summaries and preliminary report drafts from available evidence. Bedrock Guardrails and mandatory human approval workflows help ensure that AI-generated content remains subject to HSE review.
Amazon QuickSight provides operational dashboards, giving authorized personnel visibility into safety trends, incidents and compliance performance.
The result is a connected platform that moves safety monitoring beyond periodic inspections and retrospective video reviews.
Five Key Engineering Capabilities Behind the AWS Solution
1. Intelligent CCTV Video Processing with Amazon Kinesis Video Streams
The first engineering capability is the transformation of existing CCTV cameras into sources of operational safety intelligence.
Instead of transmitting every camera’s full-resolution video continuously for cloud-based analysis, the platform uses camera gateways to collect RTSP/ONVIF streams and perform initial processing locally.
The gateways apply motion and activity filtering, downscale video to 720p and prepare selected footage for cloud processing.
The architecture targets approximately 1 Mbps per camera and samples frames at one to two frames per second for inference.
Amazon Kinesis Video Streams receives the video streams and provides configurable retention of approximately 24 hours to seven days, depending on the use case. This supports the retrieval of relevant footage during incident investigation while allowing routine footage to expire automatically.
This design provides several advantages:
- Reuses Northbridge’s existing CCTV infrastructure.
- Reduces unnecessary video transmission and processing.
- Supports centralized monitoring across six facilities.
- Preserves relevant footage for incident investigation.
- Reduces bandwidth requirements compared with full-resolution continuous analysis.
The solution also uses redundant VPN connectivity and local gateway buffering to help maintain video synchronization during temporary network interruptions.
2. AI-Powered Safety Detection with Amazon SageMaker and Amazon Rekognition
Video ingestion becomes valuable when the platform can interpret what is happening within the footage.
Arthurite’s architecture uses custom machine-learning models hosted on Amazon SageMaker to identify safety-related activities and visible hazards.
The models are designed around Northbridge’s operational environment and use customer-specific camera imagery for training and evaluation.
The platform focuses on four major detection categories:
PPE compliance: Identifies potential violations involving required protective equipment, including safety helmets, high-visibility vests and safety boots.
Restricted-zone monitoring: Uses virtual boundary polygons to identify entry into controlled areas without the appropriate permit or authorization.
Smoke and flame detection: Identifies visible indicators of potential fire-related incidents and raises alerts for investigation.
Spill detection: Identifies visible liquid spills and related conditions that may require an immediate safety response.
Amazon SageMaker provides the primary custom inference capability through GPU-backed real-time endpoints. The endpoints use auto scaling across two Availability Zones, with a minimum of two instances to support continuity.
Amazon Rekognition provides an additional verification layer for flagged frames. This helps improve the quality of detections and reduce false positives.
The combined approach is important because general-purpose recognition alone may not identify every PPE requirement in a specific industrial environment. Northbridge’s requirements include items such as high-visibility vests and safety boots, which require customer-specific detection capabilities.
The architecture deliberately excludes facial recognition and individual identity identification. Its purpose is to identify safety conditions and operational risks, not to establish the identity of workers.
3. Automated Incident Detection, Alerting and Escalation
Identifying a potential hazard is only the first step. The platform must also ensure that the relevant personnel receive actionable notifications and that incidents move through a clearly defined response process.
Arthurite uses Amazon EventBridge, AWS Lambda and AWS Step Functions to coordinate incident workflows.
When the detection layer identifies a potential safety event, the platform creates a structured detection record and enriches it with relevant information.
AWS Step Functions manages the incident lifecycle, including:
- Detection and event registration.
- Alert generation and delivery.
- Incident acknowledgement.
- Assignment to the appropriate personnel.
- Response and corrective action.
- Incident closure and outcome recording.
Amazon SNS delivers SMS and other configured notifications to designated HSE personnel.
The platform also integrates with external gas-detector alarms and permit-status systems through Amazon API Gateway. This allows the solution to correlate visual observations with other operational signals.
For example, when a camera detects activity within a restricted area, the platform can compare the event with available permit information. This provides additional context for the HSE team before deciding how to respond.
Critical incidents can trigger automatic escalation if the assigned personnel do not acknowledge them within the configured response window.
The target is to deliver alerts within 30 seconds of a confirmed detection, supporting a faster response to potentially hazardous conditions.
4. Generative AI-Assisted Incident and Regulatory Reporting with Amazon Bedrock
Incident documentation can consume significant time, particularly when HSE teams must gather information from different sources and prepare reports for internal review or regulatory submission.
Arthurite integrated Amazon Bedrock into the platform to help automate parts of this process.
The solution uses Amazon Bedrock with a Claude foundation model to transform structured incident evidence into draft documentation.
The reporting capabilities include:
- Incident summaries.
- Preliminary cause analysis.
- Regulatory report drafts.
- Weekly HSE activity summaries.
- Structured incident documentation.
The platform uses available detection records, incident timelines and supporting evidence to prepare these drafts.
Amazon Bedrock Guardrails provide additional content and sensitive-information controls. Figures and structured operational values are inserted from the underlying data stores through application logic rather than relying on the language model to calculate them independently.
Most importantly, the solution maintains a mandatory human-in-the-loop approval process.
Every AI-generated incident or regulatory report must be reviewed and approved by a named HSE officer through an AWS Step Functions workflow before submission.
Nothing is submitted automatically.
If the generative AI service becomes unavailable or a request times out, the workflow allows the HSE officer to continue preparing the report manually. This ensures that AI assistance does not become a dependency that prevents essential safety documentation.
5. Centralized Safety Analytics and Operational Visibility with Amazon QuickSight
A multi-facility safety platform needs to provide more than individual notifications. Management and HSE teams also need a consistent view of safety performance across their operations.
Arthurite uses Amazon QuickSight to provide dashboards for authorized users.
The platform organizes information from detection records and incident history to support several reporting areas:
Live Safety Monitoring: Displays current safety events, camera health, alerts and incident status across the monitored facilities.
PPE Compliance: Tracks PPE observations, potential violations and compliance trends over time.
Restricted-Zone Monitoring: Provides visibility into access violations and permit-related events.
Hazard Monitoring: Displays detected smoke, flame, spill and related incidents.
Incident Management: Tracks acknowledgement, assignment, response and closure performance.
Reporting Performance: Provides visibility into incident documentation and regulatory report preparation.
AWS Glue and Amazon Athena support the preparation and analysis of historical detection data, while Amazon Cognito provides authenticated access for approximately 50 HSE users.
This centralized approach helps Northbridge move from fragmented observations and manual reporting to a more consistent operational view of safety performance.
Security, Privacy and High Availability by Design
Safety monitoring platforms process sensitive operational information. Consequently, security, privacy and reliability must be considered throughout the architecture.
Arthurite incorporated several controls into the Northbridge solution.
Encryption and key management: AWS Key Management Service (AWS KMS) customer-managed keys protect relevant video streams, storage resources, database records and Amazon Bedrock logs.
Secure connectivity: Camera gateways transmit video through redundant AWS Site-to-Site VPN tunnels. TLS protects data in transit, while user-facing dashboards and APIs use HTTPS.
Identity and access management: AWS Identity and Access Management (IAM) applies least-privilege access, while Amazon Cognito authenticates authorized HSE users.
Evidence protection: Amazon S3 Object Lock protects incident-linked evidence, and AWS CloudTrail records access to relevant evidence resources.
Application protection: AWS WAF protects the dashboard and API endpoints against common web threats.
Availability: SageMaker inference endpoints run across multiple Availability Zones with auto scaling. Amazon DynamoDB, Amazon S3 and Amazon Kinesis Video Streams support the platform’s data and evidence workflows.
Operational monitoring: Amazon CloudWatch monitors camera health, endpoint latency and other important workload indicators.
The architecture also supports privacy by design. It excludes facial recognition and individual identification, incorporates a face-blurring option and applies short retention periods to routine footage.
These measures support the organization’s privacy objectives and alignment with the Nigeria Data Protection Act 2023.
The platform remains advisory: Northbridge’s existing gas-detection systems, SCADA alarms, site safety procedures and HSE personnel remain authoritative. The AI platform supplements those controls rather than replacing them.
Expected Business Impact: What Northbridge Aims to Achieve
The Northbridge engagement establishes measurable objectives for evaluating the effectiveness of AI-powered safety monitoring.
The following table presents the documented baseline and proposed 12-month targets.
| Performance indicator | Current baseline | Proposed 12-month target |
|---|---|---|
| Working-hour coverage of monitored areas | Less than 2% | At least 95% |
| PPE non-compliance | Approximately 11% | 3% |
| Leak, smoke and spill detection-to-response | Approximately 4.5 hours | Under 10 minutes |
| High-severity alert delivery | Manual response process | Under 30 seconds |
| Incident report preparation | Approximately 6 working days | Under 1 day |
| Regulatory report preparation | Approximately 3 working days | Under 4 hours, with human approval |
| Manual inspection and reporting effort | ₦38M annually | 45% reduction |
These are project targets, not independently verified performance results. Actual KPI measurements should be added when the agreed reporting period and validation data are available.
The expected operational benefits include:
- Greater visibility across monitored working hours.
- Earlier identification of PPE violations and restricted-zone activity.
- Faster notification of visible safety hazards.
- Better preservation of incident evidence.
- Reduced administrative effort for HSE teams.
- More consistent incident and regulatory documentation.
- Improved ability to monitor safety trends across multiple facilities.
Cost Optimization and AWS Consumption Planning
The architecture was designed to balance continuous safety monitoring with the cost of processing large volumes of video.
The SOW estimates AWS consumption at approximately $5,270 per month, equivalent to $63,240 annually.
The largest estimated cost components include:
| AWS service or cost area | Estimated monthly cost |
|---|---|
| Amazon SageMaker GPU inference | $2,900 |
| Amazon Kinesis Video Streams | $540 |
| VPC and data transfer | $320 |
The project investment was estimated at ₦80 million. Managed support was priced at ₦5.2 million per month for Premium Support, which was recommended because the platform supports around-the-clock safety monitoring. Standard Support was also available at ₦2.9 million per month.
Based on the recommended Premium Support option, the estimated ongoing annual cost was approximately ₦164 million, combining AWS consumption and managed support. Site connectivity and carrier charges were excluded.
Several engineering decisions help control the ongoing cost:
- Filtering camera footage at the gateway before cloud transmission.
- Downscaling video to 720p.
- Sampling at one to two frames per second instead of analyzing every frame.
- Using auto scaling for GPU-backed SageMaker endpoints.
- Reusing the customer’s existing CCTV infrastructure.
- Applying configurable video retention rather than retaining all footage indefinitely.
GPU inference remains the largest cost driver. The SOW therefore identifies load testing and subsequent right-sizing as important activities for controlling actual consumption.
Building a More Proactive Safety Environment with AWS
Northbridge Energy’s HSE monitoring initiative demonstrates how organizations can extend existing operational infrastructure through cloud-based AI and machine learning.
By combining Amazon Kinesis Video Streams, Amazon SageMaker, Amazon Rekognition, Amazon Bedrock, AWS Step Functions and other AWS services, Arthurite Integrated implemented a platform that connects video monitoring, hazard detection, incident management and AI-assisted reporting.
More importantly, the solution places these capabilities within a controlled operational framework. AI-generated reports require human approval, existing safety systems remain authoritative, and the platform avoids facial recognition and individual identification.
For oil and gas operators managing multiple facilities, this approach offers a practical way to improve safety visibility without depending entirely on additional manual inspections or replacing existing camera infrastructure.
As industrial operations become more connected, the ability to turn operational data into timely and actionable safety intelligence will remain an important part of modern HSE management.
Arthurite Integrated continues to help organizations design and implement AWS-powered solutions that improve operational visibility, strengthen security and support better business decisions.
Looking to modernize safety monitoring across your oil and gas operations?
Arthurite Integrated can help your organization assess its existing infrastructure, design an AI-powered HSE monitoring architecture and implement cloud solutions that support safer, more efficient operations.